Examples with RowSetLoader org.apache.drill.exec.physical.rowSet.RowSetLoader used on opensource projects

Example 36 with RowSetLoader

use of org.apache.drill.exec.physical.rowSet.RowSetLoader in project drill by axbaretto.

the class TestResultSetLoaderProjection method doProjectionTest.

private void doProjectionTest(ResultSetLoader rsLoader) {
    RowSetLoader rootWriter = rsLoader.writer();
    // All columns appear, including non-projected ones.
    TupleMetadata actualSchema = rootWriter.schema();
    assertEquals(4, actualSchema.size());
    assertEquals("a", actualSchema.column(0).getName());
    assertEquals("b", actualSchema.column(1).getName());
    assertEquals("c", actualSchema.column(2).getName());
    assertEquals("d", actualSchema.column(3).getName());
    assertEquals(0, actualSchema.index("A"));
    assertEquals(3, actualSchema.index("d"));
    assertEquals(-1, actualSchema.index("e"));
    // Non-projected columns identify themselves via metadata
    assertFalse(actualSchema.metadata("a").isProjected());
    assertTrue(actualSchema.metadata("b").isProjected());
    assertTrue(actualSchema.metadata("c").isProjected());
    assertFalse(actualSchema.metadata("d").isProjected());
    // Write some data. Doesn't need much.
    rsLoader.startBatch();
    for (int i = 1; i < 3; i++) {
        rootWriter.start();
        rootWriter.scalar(0).setInt(i * 5);
        rootWriter.scalar(1).setInt(i);
        rootWriter.scalar(2).setInt(i * 10);
        rootWriter.scalar(3).setInt(i * 20);
        rootWriter.save();
    }
    // Verify. Result should only have the projected
    // columns, only if defined by the loader, in the order
    // of definition.
    BatchSchema expectedSchema = new SchemaBuilder().add("b", MinorType.INT).add("c", MinorType.INT).build();
    SingleRowSet expected = fixture.rowSetBuilder(expectedSchema).addRow(1, 10).addRow(2, 20).build();
    RowSet actual = fixture.wrap(rsLoader.harvest());
    // actual.print();
    new RowSetComparison(expected).verifyAndClearAll(actual);
    rsLoader.close();
}

Example 37 with RowSetLoader

use of org.apache.drill.exec.physical.rowSet.RowSetLoader in project drill by axbaretto.

the class TestResultSetLoaderProtocol method testBasics.

@Test
public void testBasics() {
    ResultSetLoaderImpl rsLoaderImpl = new ResultSetLoaderImpl(fixture.allocator());
    ResultSetLoader rsLoader = rsLoaderImpl;
    assertEquals(0, rsLoader.schemaVersion());
    assertEquals(ResultSetLoader.DEFAULT_ROW_COUNT, rsLoader.targetRowCount());
    assertEquals(ValueVector.MAX_BUFFER_SIZE, rsLoader.targetVectorSize());
    assertEquals(0, rsLoader.writer().rowCount());
    assertEquals(0, rsLoader.batchCount());
    assertEquals(0, rsLoader.totalRowCount());
    try {
        rsLoader.harvest();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    // Can define schema before starting the first batch.
    RowSetLoader rootWriter = rsLoader.writer();
    TupleMetadata schema = rootWriter.schema();
    assertEquals(0, schema.size());
    MaterializedField fieldA = SchemaBuilder.columnSchema("a", MinorType.INT, DataMode.REQUIRED);
    rootWriter.addColumn(fieldA);
    assertEquals(1, schema.size());
    assertTrue(fieldA.isEquivalent(schema.column(0)));
    assertSame(schema.metadata(0), schema.metadata("a"));
    try {
        rootWriter.start();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rootWriter.save();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    // Because writing is an inner loop; no checks are
    // done to ensure that writing occurs only in the proper
    // state. So, can't test setInt() in the wrong state.
    rsLoader.startBatch();
    try {
        rsLoader.startBatch();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    assertFalse(rootWriter.isFull());
    rootWriter.start();
    rootWriter.scalar(0).setInt(100);
    assertEquals(0, rootWriter.rowCount());
    assertEquals(0, rsLoader.batchCount());
    rootWriter.save();
    assertEquals(1, rootWriter.rowCount());
    assertEquals(1, rsLoader.batchCount());
    assertEquals(1, rsLoader.totalRowCount());
    // Can add a field after first row, prior rows are
    // "back-filled".
    MaterializedField fieldB = SchemaBuilder.columnSchema("b", MinorType.INT, DataMode.OPTIONAL);
    rootWriter.addColumn(fieldB);
    assertEquals(2, schema.size());
    assertTrue(fieldB.isEquivalent(schema.column(1)));
    assertSame(schema.metadata(1), schema.metadata("b"));
    rootWriter.start();
    rootWriter.scalar(0).setInt(200);
    rootWriter.scalar(1).setInt(210);
    rootWriter.save();
    assertEquals(2, rootWriter.rowCount());
    assertEquals(1, rsLoader.batchCount());
    assertEquals(2, rsLoader.totalRowCount());
    // Harvest the first batch. Version number is the number
    // of columns added.
    assertFalse(rootWriter.isFull());
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertEquals(2, rsLoader.schemaVersion());
    assertEquals(0, rootWriter.rowCount());
    assertEquals(1, rsLoader.batchCount());
    assertEquals(2, rsLoader.totalRowCount());
    SingleRowSet expected = fixture.rowSetBuilder(result.batchSchema()).addRow(100, null).addRow(200, 210).build();
    new RowSetComparison(expected).verifyAndClearAll(result);
    try {
        rootWriter.start();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rsLoader.harvest();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rootWriter.save();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    // Create a second batch
    rsLoader.startBatch();
    assertEquals(0, rootWriter.rowCount());
    assertEquals(1, rsLoader.batchCount());
    assertEquals(2, rsLoader.totalRowCount());
    rootWriter.start();
    rootWriter.scalar(0).setInt(300);
    rootWriter.scalar(1).setInt(310);
    rootWriter.save();
    assertEquals(1, rootWriter.rowCount());
    assertEquals(2, rsLoader.batchCount());
    assertEquals(3, rsLoader.totalRowCount());
    rootWriter.start();
    rootWriter.scalar(0).setInt(400);
    rootWriter.scalar(1).setInt(410);
    rootWriter.save();
    // Harvest. Schema has not changed.
    result = fixture.wrap(rsLoader.harvest());
    assertEquals(2, rsLoader.schemaVersion());
    assertEquals(0, rootWriter.rowCount());
    assertEquals(2, rsLoader.batchCount());
    assertEquals(4, rsLoader.totalRowCount());
    expected = fixture.rowSetBuilder(result.batchSchema()).addRow(300, 310).addRow(400, 410).build();
    new RowSetComparison(expected).verifyAndClearAll(result);
    // Next batch. Schema has changed.
    rsLoader.startBatch();
    rootWriter.start();
    rootWriter.scalar(0).setInt(500);
    rootWriter.scalar(1).setInt(510);
    rootWriter.addColumn(SchemaBuilder.columnSchema("c", MinorType.INT, DataMode.OPTIONAL));
    rootWriter.scalar(2).setInt(520);
    rootWriter.save();
    rootWriter.start();
    rootWriter.scalar(0).setInt(600);
    rootWriter.scalar(1).setInt(610);
    rootWriter.scalar(2).setInt(620);
    rootWriter.save();
    result = fixture.wrap(rsLoader.harvest());
    assertEquals(3, rsLoader.schemaVersion());
    expected = fixture.rowSetBuilder(result.batchSchema()).addRow(500, 510, 520).addRow(600, 610, 620).build();
    new RowSetComparison(expected).verifyAndClearAll(result);
    rsLoader.close();
    try {
        rootWriter.start();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rsLoader.writer();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rsLoader.startBatch();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rsLoader.harvest();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    try {
        rootWriter.save();
        fail();
    } catch (IllegalStateException e) {
    // Expected
    }
    // Benign to close twice
    rsLoader.close();
}

Example 38 with RowSetLoader

use of org.apache.drill.exec.physical.rowSet.RowSetLoader in project drill by axbaretto.

the class TestResultSetLoaderTorture method doTortureTest.

private void doTortureTest() {
    TupleMetadata schema = new SchemaBuilder().add("n0", MinorType.INT).addMap("m1").addNullable("n1", MinorType.INT).addMapArray("m2").addNullable("n2", MinorType.INT).addNullable("s2", MinorType.VARCHAR).addMap("m3").addNullable("n3", MinorType.INT).addArray("s3", MinorType.VARCHAR).resumeMap().resumeMap().resumeSchema().buildSchema();
    ResultSetLoaderImpl.ResultSetOptions options = new OptionBuilder().setRowCountLimit(ValueVector.MAX_ROW_COUNT).setSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    TestSetup setup = new TestSetup();
    BatchWriter batchWriter = new BatchWriter(setup, rootWriter);
    int totalRowCount = 0;
    ReadState readState = new ReadState();
    for (int batchCount = 0; batchCount < 10; batchCount++) {
        rsLoader.startBatch();
        batchWriter.writeBatch();
        // Now the hard part. Verify the above batch.
        RowSet result = fixture.wrap(rsLoader.harvest());
        // result.print();
        // Should have overflowed
        int savedCount = batchWriter.rowCount();
        assertEquals(savedCount, result.rowCount());
        totalRowCount += savedCount;
        assertEquals(totalRowCount, rsLoader.totalRowCount());
        assertEquals(batchCount + 1, rsLoader.batchCount());
        BatchReader reader = new BatchReader(setup, result.reader(), readState);
        reader.verify();
        result.clear();
    }
    // Last row overflow row
    {
        rsLoader.startBatch();
        // Use this to visualize a string buffer. There is also a method
        // to visualize offset vectors. These two are the most pesky vectors
        // to get right.
        // VectorPrinter.printStrings((VarCharVector) ((NullableVarCharVector) ((AbstractScalarWriter) batchWriter.s2Writer).vector()).getValuesVector(), 0, 8);
        RowSet result = fixture.wrap(rsLoader.harvest());
        // Use this here, or earlier, when things go amiss and you need
        // to see what the actual results might be.
        // result.print();
        totalRowCount++;
        assertEquals(totalRowCount, rsLoader.totalRowCount());
        BatchReader reader = new BatchReader(setup, result.reader(), readState);
        reader.verify();
        result.clear();
    }
    rsLoader.close();
}

Example 39 with RowSetLoader

use of org.apache.drill.exec.physical.rowSet.RowSetLoader in project drill by axbaretto.

the class TestResultSetSchemaChange method testSchemaChangeFirstBatch.

/**
 * Test the case where the schema changes in the first batch.
 * Schema changes before the first record are trivial and tested
 * elsewhere. Here we write some records, then add new columns, as a
 * JSON reader might do.
 */
@Test
public void testSchemaChangeFirstBatch() {
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator());
    RowSetLoader rootWriter = rsLoader.writer();
    rootWriter.addColumn(SchemaBuilder.columnSchema("a", MinorType.VARCHAR, DataMode.REQUIRED));
    // Create initial rows
    rsLoader.startBatch();
    int rowCount = 0;
    for (int i = 0; i < 2; i++) {
        rootWriter.start();
        rowCount++;
        rootWriter.scalar(0).setString("a_" + rowCount);
        rootWriter.save();
    }
    // Add a second column: nullable.
    rootWriter.addColumn(SchemaBuilder.columnSchema("b", MinorType.INT, DataMode.OPTIONAL));
    for (int i = 0; i < 2; i++) {
        rootWriter.start();
        rowCount++;
        rootWriter.scalar(0).setString("a_" + rowCount);
        rootWriter.scalar(1).setInt(rowCount);
        rootWriter.save();
    }
    // Add a third column. Use variable-width so that offset
    // vectors must be back-filled.
    rootWriter.addColumn(SchemaBuilder.columnSchema("c", MinorType.VARCHAR, DataMode.OPTIONAL));
    for (int i = 0; i < 2; i++) {
        rootWriter.start();
        rowCount++;
        rootWriter.scalar(0).setString("a_" + rowCount);
        rootWriter.scalar(1).setInt(rowCount);
        rootWriter.scalar(2).setString("c_" + rowCount);
        rootWriter.save();
    }
    // Fourth: Required Varchar. Previous rows are back-filled with empty strings.
    // And a required int. Back-filled with zeros.
    // May occasionally be useful. But, does have to work to prevent
    // vector corruption if some reader decides to go this route.
    rootWriter.addColumn(SchemaBuilder.columnSchema("d", MinorType.VARCHAR, DataMode.REQUIRED));
    rootWriter.addColumn(SchemaBuilder.columnSchema("e", MinorType.INT, DataMode.REQUIRED));
    for (int i = 0; i < 2; i++) {
        rootWriter.start();
        rowCount++;
        rootWriter.scalar(0).setString("a_" + rowCount);
        rootWriter.scalar(1).setInt(rowCount);
        rootWriter.scalar(2).setString("c_" + rowCount);
        rootWriter.scalar(3).setString("d_" + rowCount);
        rootWriter.scalar(4).setInt(rowCount * 10);
        rootWriter.save();
    }
    // Add an array. Now two offset vectors must be back-filled.
    rootWriter.addColumn(SchemaBuilder.columnSchema("f", MinorType.VARCHAR, DataMode.REPEATED));
    for (int i = 0; i < 2; i++) {
        rootWriter.start();
        rowCount++;
        rootWriter.scalar(0).setString("a_" + rowCount);
        rootWriter.scalar(1).setInt(rowCount);
        rootWriter.scalar(2).setString("c_" + rowCount);
        rootWriter.scalar(3).setString("d_" + rowCount);
        rootWriter.scalar(4).setInt(rowCount * 10);
        ScalarWriter arrayWriter = rootWriter.column(5).array().scalar();
        arrayWriter.setString("f_" + rowCount + "-1");
        arrayWriter.setString("f_" + rowCount + "-2");
        rootWriter.save();
    }
    // Harvest the batch and verify.
    RowSet actual = fixture.wrap(rsLoader.harvest());
    BatchSchema expectedSchema = new SchemaBuilder().add("a", MinorType.VARCHAR).addNullable("b", MinorType.INT).addNullable("c", MinorType.VARCHAR).add("d", MinorType.VARCHAR).add("e", MinorType.INT).addArray("f", MinorType.VARCHAR).build();
    SingleRowSet expected = fixture.rowSetBuilder(expectedSchema).addRow("a_1", null, null, "", 0, strArray()).addRow("a_2", null, null, "", 0, strArray()).addRow("a_3", 3, null, "", 0, strArray()).addRow("a_4", 4, null, "", 0, strArray()).addRow("a_5", 5, "c_5", "", 0, strArray()).addRow("a_6", 6, "c_6", "", 0, strArray()).addRow("a_7", 7, "c_7", "d_7", 70, strArray()).addRow("a_8", 8, "c_8", "d_8", 80, strArray()).addRow("a_9", 9, "c_9", "d_9", 90, strArray("f_9-1", "f_9-2")).addRow("a_10", 10, "c_10", "d_10", 100, strArray("f_10-1", "f_10-2")).build();
    new RowSetComparison(expected).verifyAndClearAll(actual);
    rsLoader.close();
}

Example 40 with RowSetLoader

use of org.apache.drill.exec.physical.rowSet.RowSetLoader in project drill by axbaretto.

the class TestResultSetSchemaChange method testSchemaChangeWithOverflow.

/**
 * Test a schema change on the row that overflows. If the
 * new column is added after overflow, it will appear as
 * a schema-change in the following batch. This is fine as
 * we are essentially time-shifting: pretending that the
 * overflow row was written in the next batch (which, in
 * fact, it is: that's what overflow means.)
 */
@Test
public void testSchemaChangeWithOverflow() {
    ResultSetOptions options = new OptionBuilder().setRowCountLimit(ValueVector.MAX_ROW_COUNT).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    rootWriter.addColumn(SchemaBuilder.columnSchema("a", MinorType.VARCHAR, DataMode.REQUIRED));
    rsLoader.startBatch();
    byte[] value = new byte[512];
    Arrays.fill(value, (byte) 'X');
    int count = 0;
    while (!rootWriter.isFull()) {
        rootWriter.start();
        rootWriter.scalar(0).setBytes(value, value.length);
        if (rootWriter.isFull()) {
            rootWriter.addColumn(SchemaBuilder.columnSchema("b", MinorType.INT, DataMode.OPTIONAL));
            rootWriter.scalar(1).setInt(count);
            // Add a Varchar to ensure its offset fiddling is done properly
            rootWriter.addColumn(SchemaBuilder.columnSchema("c", MinorType.VARCHAR, DataMode.OPTIONAL));
            rootWriter.scalar(2).setString("c-" + count);
            // Allow adding a required column at this point.
            // (Not intuitively obvious that this should work; we back-fill
            // with zeros.)
            rootWriter.addColumn(SchemaBuilder.columnSchema("d", MinorType.INT, DataMode.REQUIRED));
        }
        rootWriter.save();
        count++;
    }
    // Result should include only the first column.
    BatchSchema expectedSchema = new SchemaBuilder().add("a", MinorType.VARCHAR).build();
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertTrue(result.batchSchema().isEquivalent(expectedSchema));
    assertEquals(count - 1, result.rowCount());
    result.clear();
    assertEquals(1, rsLoader.schemaVersion());
    // Double check: still can add a required column after
    // starting the next batch. (No longer in overflow state.)
    rsLoader.startBatch();
    rootWriter.addColumn(SchemaBuilder.columnSchema("e", MinorType.INT, DataMode.REQUIRED));
    // Next batch should start with the overflow row, including
    // the column added at the end of the previous batch, after
    // overflow.
    result = fixture.wrap(rsLoader.harvest());
    assertEquals(5, rsLoader.schemaVersion());
    assertEquals(1, result.rowCount());
    expectedSchema = new SchemaBuilder(expectedSchema).addNullable("b", MinorType.INT).addNullable("c", MinorType.VARCHAR).add("d", MinorType.INT).add("e", MinorType.INT).build();
    assertTrue(result.batchSchema().isEquivalent(expectedSchema));
    RowSetReader reader = result.reader();
    reader.next();
    assertEquals(count - 1, reader.scalar(1).getInt());
    assertEquals("c-" + (count - 1), reader.scalar(2).getString());
    assertEquals(0, reader.scalar("d").getInt());
    assertEquals(0, reader.scalar("e").getInt());
    result.clear();
    rsLoader.close();
}